"AI Chatbots Horrible For Learning": Andrew Ng At NDTV World Summit

Artificial intelligence chatbots may be helping students finish their homework, but they could also be making them worse at learning, AI pioneer Andrew Ng warned at the NDTV World Summit, citing research that found students using AI scored 17% lower on assessments. Andrew Ng said the growing dependence on AI tools for academic work could leave students with weaker knowledge and understanding, even as their immediate performance on assignments improves.

"AI chatbots are terrible for learning," he said, describing the conclusion as controversial but supported by a growing body of evidence.

"More and more data is coming out that when students use AI, their homework scores go up. No surprise, AI does the work for you, but the final closed-book exam scores go down," he said.

Citing a study from the University of Pennsylvania and Wharton, Andrew Ng said students who used AI recorded 17% lower grades.

"If you take this 17% chatbot learning loss and compound it over the lifetime, or over many years, of a student in school, I think if we don't change that, we will be on a trajectory to graduate students that know much less," he said.

Andrew Ng, however, argued that the problem was not necessarily AI itself, but how the technology was being used in education. He said better tools could be developed to support learning, rather than simply completing tasks on behalf of students.

"The good news is, though, I think we can build much better things. It's just not the current generation of chatbots," he said.

AI Does The Homework, But Do Students Learn?

Andrew Ng's warning highlights a key challenge for schools and universities as generative AI becomes increasingly accessible. Students can use chatbots to explain concepts, solve mathematical problems, write essays and generate computer code in seconds. But producing a correct answer is not the same as understanding how to arrive at it.

When students rely on AI to complete assignments, they may receive better grades without developing the skills needed to solve similar problems independently. A closed-book examination, where students must demonstrate what they know without assistance, can reveal that gap.

The concern is particularly relevant as educational institutions grapple with how to incorporate AI into classrooms. While the technology can provide explanations, personalised practice and instant feedback, its effectiveness depends on whether it encourages students to think through problems or simply supplies answers.

Andrew Ng said there was scope to develop AI systems that could help students learn more effectively, but warned against assuming that today's chatbots were already fit for that purpose.

His comments also raise questions about how schools and universities should assess students in an era when AI can complete a growing range of academic tasks.

"I Will Not Hire A Software Engineer Who Doesn't Know How To Use AI"

Beyond the classroom, Andrew Ng said educational institutions were struggling to keep pace with the rapid transformation of the workplace.

He pointed out that many universities continued to teach curricula designed several years ago, even as AI tools were changing how professionals worked.

"Many university systems and educational systems are slow to change. All around the world, I still see universities teaching curricula from a few years ago," he said.

He said it was still possible in many places to graduate with a bachelor's degree in computer science without having used an AI coding assistant or interacted with a leading AI model through an application programming interface, or API.

"It's unfortunate that today it's still possible in many places to get a bachelor's degree in computer science without ever having used an AI coding assistant or ever made a call over the internet (called an API call) to a leading frontier AI model," Ng said.

He underlined how dramatically expectations in the technology industry had changed.

"Today, I will not hire a software engineer who doesn't know how to use an AI coding agent," he said.

Ng acknowledged that reforming education systems was difficult but argued that it was necessary to ensure graduates were equipped for the demands of the modern workplace.

"We do need to change it, though, to make sure that we help our students ready themselves for the job today and the future," he said.

Andrew Ng agreed that much more needed to be done worldwide to prepare people for an AI-enabled economy.